import os
import json
import traceback
import spaces

import gradio as gr
import torch
from torchvision import transforms
from huggingface_hub import hf_hub_download


os.environ.setdefault("OMP_NUM_THREADS", "1")
torch.set_num_threads(1)

REPO_ID = "Pablogps/kedar-15-birds"


# Download model and labels
model_path = hf_hub_download(
    repo_id=REPO_ID,
    filename="model.pt"
)

labels_path = hf_hub_download(
    repo_id=REPO_ID,
    filename="labels.json"
)

with open(labels_path, "r") as f:
    labels = json.load(f)


# Load TorchScript model
model = torch.jit.load(model_path, map_location="cpu")
model.eval()
model.to("cuda")


# Inference preprocessing
preprocess = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225],
    ),
])


@spaces.GPU
def predict(img):
    try:
        if img is None:
            return {}

        img = img.convert("RGB")

        x = preprocess(img).unsqueeze(0).to("cuda")

        with torch.inference_mode():
            logits = model(x)
            probs = torch.softmax(logits, dim=1)[0]

        return {
            labels[i]: float(probs[i])
            for i in range(len(labels))
        }

    except Exception as e:
        tb = traceback.format_exc()
        print(tb, flush=True)
        raise gr.Error(f"{type(e).__name__}: {e}")


title = "Bird classifier"

description = (
    "A small model that classifies birds, trained quickly "
    "with a subset of kedar 200 using 15 classes"
)

examples = [
    "examples_american_crow.jpg",
    "examples_barn_swallow.jpg",
    "examples_pied_kingfisher.jpg",
]


demo = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil"),
    outputs=gr.Label(num_top_classes=3),
    title=title,
    description=description,
    examples=examples,
    cache_examples=False,
)

demo.queue(max_size=8).launch(
    show_error=True,
    debug=True
)